Alessandra Pedrocchi

Despite a decade in, immunotherapy (IO) treatment selection in non-small cell lung cancer (NSCLC) remains largely guided by subgroup analyses and imperfect programmed death ligand 1 (PD-L1) and clinical scores. To our knowledge, I ³ LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients. We integrated real-world clinical and blood (CB) data, computed tomography (CT) images, digital pathology (DP), and genomics into machine learning early fusion (MLEF) and deep learning intermediate fusion (DLIF) models. Machine learning (ML) and deep learning (DL) CB-only models achieved consistent performance across outcomes with area under the curve (AUC) up to 0.77 in the test (TEST) set. Performance drop in external validation (EXVAL) likely reflects population differences (AUC range: 0.55–0.72). AI models significantly surpassed PD-L1, Eastern Cooperative Oncology Group performance status (ECOG PS), neutrophil-to-lymphocyte ratio (NLR), lactate dehydrogenase (LDH) and Lung Immune Prognostic Index (LIPI) score in the independent TEST set. The clinical usability study showed that lung expert and nonexpert physicians improved their prediction with the explainable AI (XAI) ML CB-only based tool. Although multimodal integration with MLEF (CB+CT+DP) was associated with higher performance, its incremental benefit remains uncertain, not translated in TEST and EXVAL. The I ³ LUNG project is a pioneering framework showing the clinical usefulness of AI tools. A prospective validation of the decision support system (both CB and multimodal) is currently undergoing in more than 2,000 patients.

Assessment of upper-limb kinematics is essential in clinical practice for diagnosis, rehabilitation monitoring, and treatment personalization. Markerless Motion Capture (MMC) systems, such as the Microsoft Azure Kinect (AK), offer a low-cost and time-efficient alternative to marker-based systems. However, while AK accuracy has been extensively studied for lower-limb movements, its performance for upper-limb analysis—especially under clinically relevant, suboptimal conditions—remains underexplored. This study aims to validate AK for upper-limb motion tracking against a gold-standard optoelectronic system under optimal and suboptimal conditions. Sixteen healthy adults performed ten upper body motor tasks in three scenarios: optimal setup, seated posture with table occlusion, and increased camera distance. Joint angles were compared using normalized Root Mean Squared Error (nRMSE) and Pearson’s correlation coefficient. Performance Indicators (PIs) including Range of Motion (ROM), smoothness, and Time to Peak Velocity (TTPV) were also evaluated. AK accurately captured movements performed within the camera plane, with median nRMSE below 20% in optimal conditions and no significant degradation in suboptimal setups. In contrast, movements occurring on planes perpendicular to the camera were poorly captured. ROM estimation was acceptable and highly reproducible, while TTPV showed moderate-to-poor reliability and smoothness deviated substantially from the reference system. These findings suggest that careful attention to Kinect positioning is essential to ensure effective acquisitions, even in suboptimal scenarios. Future research should evaluate AK validity in clinical populations and explore the effects of system interference in multi-device setups.

Parkinson's disease is defined by dopaminergic neuron loss in the substantia nigra, yet its hallmark, exaggerated beta-band synchrony, pervades motor cortex, thalamus, and cerebellum, implicating network dynamics far beyond any single circuit. How focal subcortical dopamine depletion translates into brain-wide oscillatory pathology remains unresolved. We use a connectome-constrained multiscale model of the mouse brain, embedding biophysically detailed spiking networks of basal ganglia and cerebellum within whole-brain corticothalamic dynamics grounded in the Allen Mouse Brain Connectivity Atlas. We show that confining dopamine depletion exclusively to subcortical circuits is sufficient to produce widespread beta hypersynchrony (10-30 Hz), accompanied by heterogeneous theta and gamma dysregulation. Virtual loop ablations reveal that cortical and cerebellar beta amplification strictly requires intact cortico-basal ganglia-thalamic feedback; severing this loop confines beta to subcortical generators. These results support resonance within closed large-scale loops, rather than local rhythmogenesis, as the mechanism underlying distributed Parkinsonian beta pathology.

The thalamus plays a crucial role in motor control but its complex circuitry remains poorly understood. Existing computational models often lack anatomical detail, hindering investigations on how structure influences function. Here we present a data-driven 3D anatomical scaffold model (a geometrically constrained virtual circuit) of mouse motor thalamic nuclei, built by integrating publicly available datasets, anatomical descriptions, geometric constraints and circuit-level findings to reproduce topographical organization and structural boundaries observed experimentally. Network simulations demonstrate sustained spindle oscillations at physiologically realistic frequencies, with propagation velocities matching observations from thalamic slices. Systematic ablation reveals that both topography and distance-dependent synaptic weights are necessary for physiological dynamics, establishing architectural design principles for spatially organized circuits. These findings generalize beyond the motor thalamus: the open-source pipeline provides a reusable framework for data-driven circuit reconstruction, linking anatomical organisation to emergent network dynamics across brain regions.

Reduced shoulder mobility limits upper-limb function and daily activities across a wide range of conditions. Wearable exosuits have shown promise in assisting arm movements and reducing muscle effort; however, comfort has rarely been prioritized as an explicit design objective, despite it strongly affects mass-adoption and long-term usage. This study presents a redesigned soft shoulder exosuit developed to address comfort-related limitations identified in our previous version. In parallel, the actuator-harness configuration has been redesigned to bias assistance preferentially toward shoulder flexion, to enhance functionally relevant hand positioning. A comparison between the previous (v1) and redesigned (v2) versions was conducted in eight healthy participants, who performed static, dynamic, and functional tasks, with muscle activity, kinematics, and user-reported feedback as main outcomes. Both versions increased endurance time and reduced deltoid activation when assistance was active, without hindering movement freedom when unpowered. However, the difference emerged most clearly during functional tasks and comfort evaluation. Version 2 was associated with forward-directed arm elevation and larger transverse-plane mobility by up to 30°, without a detected increase in deltoid activation. User-reported outcomes further suggested improved wearability, with lower perceived pressure across the harness and higher ratings in effectiveness (+23%), ease of use (+20%), and comfort (+17%) compared to the previous version. Taken together, these findings suggest that redesign strategies focused on wearability and task-oriented assistance can improve comfort and functional interaction without compromising assistive performance.

Neurorehabilitation conventionally relies on the interaction between a patient and a physical therapist. Robotic systems can improve and enrich the physical feedback provided to patients after neurological injury, but they under-utilize the adaptability and clinical expertise of trained therapists. In this position paper, we advocate for a novel approach that integrates the therapist’s clinical expertise and nuanced decision-making with the strength, accuracy, and repeatability of robotics: Robot-mediated physical Human–Human Interaction. This framework, which enables two individuals to physically interact through robotic devices, has been studied across diverse research groups and has recently emerged as a promising link between conventional manual therapy and rehabilitation robotics, harmonizing the strengths of both approaches. Although current findings are largely based on pilot studies and conceptual frameworks, integrating therapists’ expertise with the functionalities offered by robotic systems represents a promising direction for improving rehabilitation outcomes. This paper presents the rationale of a multidisciplinary team—including engineers, doctors, and physical therapists—for conducting research that utilizes: a unified taxonomy to describe robot-mediated rehabilitation, a framework of interaction based on social psychology, and a technological approach that makes robotic systems seamless facilitators of natural human-human interaction.

Real-world data (RWD), largely stored in unstructured electronic health records (EHRs), are critical for understanding complex diseases like cancer. However, extracting structured information from these narratives is challenging due to linguistic variability, semantic complexity, and privacy concerns. This study evaluates the performance of four locally deployable and small language models (SLMs), LLaMA, Mistral, BioMistral, and MedLLaMA, for information extraction (IE) from Italian EHRs within the APOLLO 11 trial on non-small cell lung cancer (NSCLC). We examined three prompting strategies (zero-shot, few-shot, and annotated few-shot) across English and Italian, involving clinicians with varying expertise to assess prompt design’s impact on accuracy. Results show that general-purpose models (e.g., LLaMA 3.1 8B) outperform biomedical models in most tasks, particularly in extracting binary features. Multiclass variables such as TNM staging, PD-L1, and ECOG were more difficult due to implicit language and lack of standardization. Few-shot prompting and native-language inputs significantly improved performance and reduced hallucinations. Clinical expertise enhanced consistency in annotation, particularly among students using annotated examples. The study confirms that privacy-preserving SLMs can be deployed locally for efficient and secure cancer data extraction. Findings highlight the need for hybrid systems combining SLMs with expert input and underline the importance of aligning clinical documentation practices with SLM capabilities. This is the first study to benchmark SLMs on Italian EHRs and investigate the role of clinical expertise in prompt engineering, offering valuable insights for the future integration of SLMs into real-world clinical workflows.

The integration of robotics and Electrical Stimulation (ES) in neurorehabilitation leverages robotics’ precise task execution alongside ES-induced motor learning, muscle conditioning, and cardiovascular benefits. We propose a hybrid system for overground gait training, combining neuromuscular ES and a motorized exoskeleton. Different combination modalities are proposed: ES-motor cooperation for the swinging knee, synchronized but independent ES and motor assistance for hip movements and for the knee during stance, and ES-only for the non-actuated ankle. Twelve non-disabled subjects and eleven participants with neurological disorders tested the system under two conditions: exoskeleton-only and hybrid. The hybrid condition reduced knee motor torque by 48% during swing without compromising tracking accuracy, showing that ES can effectively drive limb motion. Neurological participants rated the hybrid system as more usable than the exoskeleton alone (median 5-point improvement of System Usability Scale). These findings support the feasibility of hybrid ES-motorized exoskeletons in clinical settings. Future studies should investigate their potential to enhance therapeutic outcomes.

Sound localization relies on precise processing of binaural cues in medial (MSO) and lateral superior olive (LSO). However, key questions remain on how these two nuclei perform their specific computations depending on sound frequency, the relative contributions of interaural time differences (ITDs) and interaural level differences (ILDs), as well as the role of inhibitory timings. Experimental studies have struggled to address these issues because of technical challenges and a lack of methodological consistency. Here, we present a comprehensive computational model of auditory peripheral and brainstem neural populations to investigate how ITDs and ILDs are encoded by LSO and MSO. We developed a spiking neural network with realistic tonotopic organization and biologically consistent synaptic connections. We tested responses to pure tones and white noise from different locations under three cue conditions: human-recorded head-related transfer functions, isolated ITDs, and isolated ILDs. LSO neurons showed realistic ipsilateral-preferring responses across different stimuli, with cue dependency varying by tone frequency, while white noise responses were driven mostly by ILDs. MSO responses showed heterogeneous tuning, with contralateral preference for low-frequency tones, which got lost for higher-frequency ones, and white noise response driven by ITDs. We further validated two experimental findings: (1) removing MSO inhibition abolished contralateral tuning, and (2) varying the timing between excitation and inhibition produced large shifts in tuning, highlighting the importance of synaptic timing for ITD coding. This model serves as an in silico testbed for auditory research, offering new insights into the functioning of human spatial hearing.

OBJECTIVE: Reaching training with error augmentation (EA) has recently shown great promise for enhancing bimanual therapeutic training, using both robotic forces feedback (haptics) and a visually distorted display elements (graphics) to amplify motor learning. METHODS: Here in a two-arm, randomized controlled trial we explored the effect of visual EA alone by visually shifting the paretic limb’s cursor in the direction of error. We invited 38 chronic (> 8 months post injury) stroke survivors to practice bimanual reaching for approximately 40 minutes, 3 days per week, for three weeks. RESULTS: Arm motor section of the Fugl-Meyer (AMFM; maximum score 66 points) increased an average of 2.2 and retained to a follow-up evaluation 7-9 weeks (about 2 months) later (average 1.5). Clinically meaningful increase for AMFM for chronic stroke survivors is 5.2 points. No superiority was detected due to the EA treatment, but other measures on the composite abilities (range of motion, bimanual symmetry, and movement time) showed improvements favoring EA. CONCLUSIONS: While removing robot forces led to smaller gains than previous work, such touch-free bimanual therapy may still prove to be an effective inexpensive automated rehabilitation tool for wider accessibility in therapy interventions. This study was registered at ClinicalTrials.gov (ID# NCT03300141 ).